SuSha: A multi-model ensemble learning framework for predicting microbial salinity adaptation

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Microbial salinity adaptation research is hindered by single-gene model limitations and challenges in studying complex ecosystems.
• Understanding systemic responses to salinity stress in natural habitats requires advanced computational approaches.

Purpose of the Study:

• To develop a robust computational framework, SuSha, for predicting microbial salinity adaptation using genome-wide features.
• To overcome the limitations of traditional models by leveraging ensemble learning and whole-genome data.

Main Methods:

• Extracted a 24-dimensional feature vector from whole-genome data of 123 bacterial and archaeal species, including amino acid frequencies.
• Developed an ensemble model integrating random forest, bagging, and extra trees algorithms.
• Validated the model using five-fold cross-validation and external testing on 2678 metagenomic samples from diverse global habitats.

Main Results:

• The ensemble model achieved a global accuracy of 0.765 and an AUC of 0.941, significantly outperforming individual models.
• SuSha demonstrated high robustness and ecological consistency across salinity gradients.
• Achieved over 90% classification accuracy for extreme halophiles, particularly in hypersaline environments.

Conclusions:

• SuSha provides a powerful computational tool for high-precision genotype-to-phenotype predictions of microbial salinity adaptation.
• The framework enables inference of physiological potential for uncultivated microorganisms and facilitates resource mining in extreme environments.